Test Stinky

A holistic quality detector and gate stack for test suites, in any language and any framework. It exists as a counterweight, agents love to cover code with tests and overshoot: they ship suites that are huge, slow, and flaky, where coverage is the goal instead of a map, and where a green badge stops meaning anything. It finds the patterns that make a suite expensive to run, impossible to trust, and painful to change, explains the cost of each, and proposes a concrete fix. The full catalog with detection signals, fixes, exceptions, and sources is in catalog.md; read it before running a scan.

The judged artifact is the tests, not the production code. Production smells found along the way defer to sibling skills (react-stinky for React/TypeScript, tauri-stinky for Rust and Tauri); if those are not installed, note the finding in one line and move on. Everything about the tests themselves, the fixtures, the doubles, the waiting, the lanes, and the CI hygiene around them is in scope.

What it sniffs for

Nine pillars, 49 categories. Detection signals, fixes, and sources live in catalog.md. The background models the findings lean on (the test pyramid, the doubles taxonomy, flakiness, hermeticity, DAMP vs DRY, mutation and property-based testing) are separate linked concepts under references/concepts/; read one when a finding needs the underlying model explained, not just named.

1. Test intent and shape. Vague names, eager multi-behavior tests, logic in tests, structure mirroring and internals-poking, god test files, rotting disabled tests.
2. Assertions. Assertion-free tests, weak asserts where exact values are known, tautologies, snapshot-everything, wide brittle asserts, assertion roulette, implementation-detail asserts.
3. Fixtures, setup, and duplication. Mystery guests, general fixtures, over-DRY abstraction (DAMP wins), copy-paste setup drifting across files, irrelevant-detail noise.
4. Test doubles. Over-mocking, mocking what you don't own, mock drift with no contract test, deep mock chains, partial mocks of the unit under test.
5. Determinism. Real clocks, unseeded randomness, order dependence, shared mutable state, unhermetic units, platform and locale dependence, concurrency races.
6. Async and waiting. Sleep-based synchronization, missing awaits, unbounded or globally-inflated waits, retry loops inside test bodies.
7. Speed and cost. Inverted pyramid, heavy per-test setup, the same behavior tested at three layers, forced serial execution, mega-parametrization, no speed budget.
8. Coverage theater and gaps. Coverage-driven tests, trivial tests, happy-path-only suites, tests that cannot fail.
9. Lifecycle and CI hygiene. Retry as policy, quarantine without a ledger, green-by-rerun culture, silent conditional passes, asserts fixed by deletion.